A regularization technique that penalizes large parameter values during training to prevent overfitting.